03-Coolidge-4857.

qxd

1/2/2006

5:43 PM

Page 67

Statistical Parameters
Measures of Central Tendency and Variation

Chapter 3 Goals
Understand and compute measures of central tendency Understand and compute measures of variation Learn the differences between klinkers and outliers and how to deal with them Learn how Tchebysheffs theorem led to the development of the empirical rule Learn about the sampling distribution of means, the central limit theorem, and the standard error of the mean Use the empirical rule for prediction

__________________________ Measures of Central Tendency

In addition to graphs and tables of numbers, statisticians often use common parameters to describe sets of numbers. There are two major categories of these parameters. One group of parameters measures how a set of numbers is centered around a particular point on a line scale or, in other words, where (around what value) the numbers bunch together. This category of parameters
67

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 68

68

STATISTICS: A GENTLE INTRODUCTION

is called measures of central tendency. You already know and have used the most famous statistical parameter from this category, which is the mean or average.

The Mean
The mean is the arithmetic average of a set of scores. There are actually different kinds of means, such as the harmonic mean (which will be discussed later in the book) and the geometric mean. We will first deal with the arithmetic mean. The mean gives someone an idea where the center lies for a set of scores. The arithmetic mean is obtained by taking the sum of all the numbers in the set and dividing by the total number of scores in the set. You already learned how to derive the mean in Chapter 1. Here again is the official formula in proper statistical notation:
k

x =

i=1

xi ,

where - = the mean, x

k i=1

xi = the sum of all the scores in the set,

N = the number of scores or observations in the set. Note that from this point on in the text, we will use the shorthand k symbol x instead of the proper xi.
i=1

The mean has many important properties that make it useful. Probably its most attractive quality is that it has a clear conceptual meaning. People almost automatically understand and easily form a picture of an unseen set of numbers when the mean of that set is presented alone. Another attractive quality is that the mathematical formula is simple and easy. It involves only adding, counting, and dividing. The mean also has some more complicated mathematical properties that also make it highly useful in more advanced statistical settings such as inferential statistics. One of these properties is that the mean of a sample is said to be an unbiased estimator of the population mean. But first, let us back up a bit. The branch of statistics known as inferential statistics involves making inferences or guesses from a sample about a population. As members of society, we continually make decisions, some big (such as what school to attend, who to marry, or whether to have an operation) and some little (such as what clothes to wear or what brand of soda to buy). We hope that our big decisions are based on sound research. For example, if we decide to take a drug to lower high

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 69

Chapter 3: Statistical Parameters

69

blood pressure, we hope that the mean response of the participants to the drug is not just true of the sample but also of the population, that is, all people who could take the drug for high blood pressure. So, if we take the mean blood pressure of a sample of patients after taking the drug, we hope that it will serve as an unbiased estimator of the population mean; that is, the mean of a sample should have no tendency to overestimate or underestimate the population mean, (which is also written mu and pronounced mew). Thus, if consecutive random samples are drawn from a larger population of numbers, each sample mean is just as likely to be above as it is to be below . This property is also useful because it means that the population formula for is the same as the -. These formulas are as follows: sample formula for x

Sample Mean
= x x N

Population
= x N

The Median
Please think back to when you heard income or wealth reports in the United States. Can you recall if they reported the mean or the median income (or wealth)? More than likely, you heard reports about the median and not the mean. Although the mean is the mostly widely used measure of central tendency, it is not always appropriate to use it. There may be many situations where the median may be a better measure of central tendency. The median value in a set of numbers is that value that divides the set into equal halves when all the numbers have been ordered from lowest to highest. Thus, when the median value has been derived, half of all the numbers in the set should be above that score, and half should be below that score. The reason that the median is used in reports about income or wealth in the United States is that income and wealth are unevenly distributed (i.e., they are not normally distributed). A plot of the frequency distribution of income or wealth would reveal a skewed distribution. Now, would the resulting distribution be positively skewed or negatively skewed? It is easier to answer that question if you think in terms of wealth. Are most people in the United States wealthy and theres just a few outlying poor people, or do most people have modest wealth and a few people own a huge amount? Of course, the latter is true. It is often claimed that the wealthiest 3% of U.S. citizens own about 40% of the total wealth. Thus, wealth is positively skewed. A mean value of wealth would be skewed or drawn to the right by a few higher values, giving the appearance that the average person in the United States is far wealthier than he or she really is. In all skewed

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 70

70

STATISTICS: A GENTLE INTRODUCTION

distributions, the mean is strongly influenced by the few high or low scores. Thus, the mean may not truthfully represent the central tendency of the set of scores because it has been raised (or lowered) by a few outlying scores. In these cases, the median would be a better measure of central tendency. The formula for obtaining the median for a set of scores will vary depending on the nature of the ordered set of scores. The following two methods can be used in many situations.

Method 1
When the scores are ordered from lowest to highest and there are an odd number of scores, the middle value will be the median score. For example, examine the following set of scores: 7, 9, 12, 13, 17, 20, 22 Because there are seven scores and 7 is an odd number, then the middle score will be the median value. Thus, 13 is the median score. To check whether this is true, look to see whether there is exactly the same number of scores above and below 13. In this case, there are three scores above 13 (17, 20, and 22) and three scores below 13 (7, 9, and 12).

Method 2
When the scores are ordered from lowest to highest and there are an even number of scores, the point midway between the two middle values will be the median score. For example, examine the following set of scores: 2, 3, 5, 6, 8, 10 There are six scores, and 6 is an even number; therefore, take the average of 5 and 6, which is 5.5, and that will be the median value. Notice that in this case, the median value is a hypothetical number that is not in the set of numbers. Let us change the previous set of numbers slightly and find the median: 2, 3, 5, 7, 8, 10 In this case, 5 and 7 are the two middle values, and their average is 6; therefore, the median in this set of scores is 6. Let us obtain the mean for this last set, and that is 5.8. In this set, the median is actually slightly higher than the mean. Overall, however, there is not much of a difference between these two measures of central tendency. The reason for this is that the numbers are relatively evenly distributed in the set. If the population from which this

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 71

Chapter 3: Statistical Parameters

71

sample was drawn is normally distributed (and not skewed), then the mean and the median of the sample will be about the same value. In a perfectly normally distributed sample, the mean and the median will be exactly the same value. Now, let us change the last set of numbers once again: 2, 3, 5, 7, 8, 29 Now the mean for this set of numbers is 9.0, and the median remains 6. Notice that the mean value was skewed toward the single highest value (29), while the median value was not affected at all by the skewed value. The mean in this case is not a good measure of central tendency because five of the six numbers in the set fall below the mean of 9.0. Thus, the median may be a better measure of central tendency if the set of numbers has a skewed distribution. See Figure 3.1 for graphic examples.

Frequency

Mean

Median

Frequency

Median

Mean

Figure 3.1

If there are ties at the median value when you use either of the two previous methods, then you should consult an advanced statistics text for a third median formula, which is much more complicated than the previous two methods. For example, examine the following set: 2, 3, 5, 5, 5, 10 There are an even number of scores in this set, and normally we would take the average of the two middle values. However, there are three 5s, and that constitutes a tie at the median value. Notice that if we used 5 as the value of the median, there is one score above the value 5, and there are two scores below 5. Therefore, 5 is not the correct median value.

The Mode
The mode is a third measure of central tendency. The mode score is the most frequently occurring number in a set of scores. In the previous set of

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 72

72

STATISTICS: A GENTLE INTRODUCTION

numbers, 5 would be the mode score because it occurs at a greater frequency than any other number in that set. Notice that the mode score in that set is 5, and the frequency (how many are there) of the mode score is 3 because there are three 5s in the set. It is also possible to have two or more mode scores in a set of numbers. For example, examine this set: 2, 3, 3, 3, 4, 5, 6, 6, 6, 8 In this set, there are two modes: One mode score is 3, and the other mode score is 6. The frequency of both mode scores is 3. A distribution that has two different modes is said to be bimodally distributed. The mode score can change drastically across different samples, and thus it is not a particularly good overall measure of central tendency. The mode probably has its greatest value as a measure with nominal or categorical scales. For example, we might report in a study that there were 18 male and 14 female participants. Although it may appear obvious and highly intuitive, we know that there were more male than female participants because we can see that 18 (males) is the mode score.

Choosing Between Measures of Central Tendency ________________________

In one of my journal articles, Dreams of the Dying (Coolidge & Fish, 1983), an undergraduate student and I obtained dream reports from 14 dying cancer patients. In the method section of the article, we reported the subjects ages. Rather than list the ages for all 14 subjects, what category of statistical parameters would be appropriate? Of course, it would be the category of measures of central tendency. The following numbers represent the subjects ages at the time of their deaths: 28, 34, 40, 40, 42, 43, 45, 48, 59, 59, 63, 63, 81, 88 The mean for this set of scores is 52.4, and the median is 46.5 [(45 + 48)/2]. Typically, a researcher would not report both the mean and the median, so which of the two measures would be reported? A graph of the frequency distribution (by intervals of 10 years) shows that the distribution appears to be skewed right. See Figure 3.2 for two versions of the frequency distribution. Because of this obvious skew, the mean is being pulled by the extreme scores of 81 and 88. Thus, in this case, we reported the median age of the subjects instead of the mean. In most statistical situations, the mean is the most commonly used measure of central tendency. Besides its ability to be algebraically manipulated,

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 73

Chapter 3: Statistical Parameters

73

6 5 4 3 2 1 2029 4049 6069 8089 3039 5059 7079 Age Intervals

6 Frequency 5 4 3 2 1 2034 5064 8094 3549 6579 Age Intervals

Figure 3.2

which allows it to be used in conjunction with other statistical formulas and procedures, the mean also is resistant to variations across different samples.

________________________________ Klinkers and Outliers

Sometimes when we are gathering data, we may have equipment failure, or in a consumer preference study, we may have a subject who does not speak English. A datum from these situations may be simply wrong or clearly inappropriate. Abelson (1995) labels these numbers klinkers. Abelson argues that when it is clearly inappropriate to keep klinkers, they should be thrown out, and additional data should be gathered in their place. Can I do this? Is this ethical? You may legitimately ask yourself these questions. Tukey (1969) and Abelson (1995) both warn against becoming too stuffy and against the sanctification of statistical rules. If equipment failure has clearly led to an aberrant score, or if some participants in a survey failed to answer some questions because they did not speak English, then our course is clear. Keeping these data would ruin the true spirit of the statistical investigative process. We are not being statistically conservative but foolish if we kept such data. The other type of aberrant score in our data is called an outlier. Outliers are deviant scores that have been legitimately gathered and are not due to equipment failures. While we may legitimately throw out klinkers, outliers are a much murkier issue. For example, the sports section of a newspaper reported that the Chicago Bulls were the highest paid team (19961997) in the National Basketball Association, with an average salary of $4.48 million. However, lets examine the salaries and see if the mean is an accurate measure of the 13 Bulls players. As a measure of central tendency, the mean salary is misleading. Only two players have salaries above the mean, Jordan and Rodman, while 11 players are below the mean. A better measure of central tendency for these data would be the median. Randy Browns salary of $1,300,000 would be the median salary since 6 players have salaries above that number and 6 players

Frequency

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 74

74 Table 3.1 Player Jordan Rodman Kukoc Harper Longley Pippen Brown Simpkins Parish Wennington Kerr Caffey Buechler

STATISTICS: A GENTLE INTRODUCTION

Salary $30,140,000 $9,000,000 $3,960,000 $3,840,000 $2,790,000 $2,250,000 $1,300,000 $1,040,000 $1,000,000 $1,000,000 $750,000 $700,000 $500,000

are below that number. It is clear that Michael Jordans salary is skewing the mean, and his datum would probably be considered an outlier. Without Jordans salary, the Bullss average pay would be about $2,300,000, which is less than half of what the Bullss average salary would be with his salary. Furthermore, to show how Jordans salary skews the mean, Michael Jordan made about $3,000,000 more than the whole rest of the teams salaries combined! We will revisit the issue of outliers later in this chapter. Notice in Jordans case, we did not eliminate his datum as an outlier. First, we identified it as an outlier (because of its effect of skewing the mean of the distribution) and reported the median salary instead. Second, we analyzed the data (team salaries) both ways, with Jordans salary and without, and we reported both statistical analyses. There is no pat answer to dealing with outliers. However, the latter approach (analyzing the data both ways and reporting both of the analyses) may be considered semi-conservative. The more conservative position would be to analyze the data and report them with the outlier and have no other alternative analyses without the outlier.

Uncertain or Equivocal Results ________________________

One of my colleagues sums up statistics to a single phrase, its about measurement. Often, as in the case of klinkers and outliers, measurement is not as straightforward as it would appear. Sometimes measurement issues arise because there is a problem with the measurement procedures or measurement. When the results of the measurement are unclear, uncertain, or unreadable, they are said to be equivocal. Intermediate results occur when a test outcome is neither positive nor negative but falls between these two conditions. An example might be a home

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 75

Chapter 3: Statistical Parameters

75

pregnancy kit that turns red if a woman is pregnant and stays white if she is not. If it turns pink, how are the results to be interpreted? Indeterminate results may occur when a test outcome is neither positive nor negative and does not fall between these two conditions. Perhaps, in the pregnancy kit example, this might occur if the kit turns green. Finally, it may be said that uninterpretable results have occurred when a test has not been given according to its correct instructions or directions, or when it yields values that are completely out of range.

_________________________________ Measures of Variation

The second major category of statistical parameters is measures of variation. Measures of variation tell us how far the numbers are scattered about the center value of the set. They are also called measures of dispersion. There are three common parameters of variation: the range, standard deviation, and variance. While measures of central tendency are indispensable in statistics, measures of variation provide another important yet different picture of a distribution of numbers. For example, have you heard of the warning against trying to swim across a lake that averages only 3 feet deep? While the mean does give a picture that the lake on the whole is shallow, we intuitively know that there is danger because while the lake may average 3 feet in depth, there may be much deeper places as well as much shallower places. Thus, measures of central tendency are useful in understanding how scores cluster about a center value, and measures of variation are useful in understanding how far, wide, or deep the high scores are scattered about the center value.

The Range
The range is the simplest of the measures of variation. The range describes the difference between the lowest score and the highest score in a set of numbers. Typically, statisticians do not actually report the range value, but they do state the lowest and highest scores. For example, given the set of scores 85, 90, 92, 98, 100, 110, 122, the range value is 122 85 = 37. Therefore, the range value for this set of scores is 37, the lowest score is 85, and the highest score is 122. It is also important to note that the mean for this set is 99.6, and the median score is 98. However, neither of these measures of central tendency tells us how far the numbers are from this center value. The set of numbers 95, 96, 97, 98, 99, 103, 109 would also have a mean of 99.6 and a median of 98. However, notice how the range, as a measure of variation, tells us that in the first set of numbers, they are widely distributed about the center (range = 37), while in the second set, the range is only 14 (109 95 = 14). Thus, the second set varies less about its center than the first set.

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 76

76

STATISTICS: A GENTLE INTRODUCTION

Let us refer back to the ages of the dream subjects in the previously mentioned study. Although the mean of the subjects ages was 52.4, we have no idea how the ages are distributed about the mean. In fact, because the median was reported, the reader may even suspect that the ages are not evenly distributed about the center. The reader might correctly guess that the ages might be skewed, but the reader would not know whether the ages were positively or negatively skewed. In this case, the range might be useful. For example, it might be reported that the scores ranged from a low of 28 to a high of 88 (although the range value itself, which is 60, might be of little conceptual use). Although the range is useful, the other two measures of variation, standard deviation and variance, are used far more frequently. The range is useful as a preliminary descriptive statistic, but it is not useful in more complicated statistical procedures, and it varies too much as a function of sample size (the range goes up when the sample size goes up). The range also depends only on the highest and lowest scores (all of the other scores in the set do not matter), and a single aberrant score can affect the range dramatically.

The Standard Deviation

The standard deviation is a veritable bulwark in the sea of statistics. Along with the mean, the standard deviation is a theoretical cornerstone in inferential statistics. The standard deviation gives an approximate picture of the average amount each number in a set varies from the center value. To appreciate the standard deviation, let us work with the idea of the average deviation. Let us work with a small subsample of the ages of the dream subjects: 28, 42, 48, 59, 63 Their mean is 48.0. Let us see how far each number is from the mean. Note that the positive and negative signs tell us whether an individual number is above or below the mean. The size or magnitude of the distance score tells us how far that number is from the mean.
-) Mean (x 48 48 48 48 48 -) Distance from Mean (xi x

Each Number (xi) 28 42 48 59 63

20 6 0 +11 +15

To get the average deviation for this set of scores, we would normally sum the five distance values and divide by 5 because there were five scores. In this case, however, if we sum 20, 6, 0, +11, and +15, we would get 0 (zero), and 0 divided by 5 is 0.

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 77

Chapter 3: Statistical Parameters

77

One solution to this dilemma would be to take the absolute value of each distance. This means that we would ignore the negative signs. If we now try to average the absolute values of the distances, we would obtain the sum of 20, 6, 0, 11, and 15, which is 52, and 52 divided by 5 is 10.4. Now, we have a picture of the average amount each number varies from the mean, and that number is 10.4. The average of the absolute values of the deviations has been used as a measure of variation, but statisticians have preferred the standard deviation as a better measure of variation, particularly in inferential statistics. One reason for this preference, especially among mathematicians, is that the absolute value formula cannot be manipulated algebraically. The formula for the standard deviation for a population is
(xi x )2 N

Note that or sigma represents the population value of the standard devias the command to sum numbers together. ation. You previously learned is the capital Greek letter, and is the lowercase Greek letter. Also note that although they are pronounced the same, they have radically different meanings. The sign is actually in the imperative mode; that is, it states a command (to sum a group of numbers). The other value is in the declarative mode; that is, it states a fact (it represents the population value of the standard deviation). The sample standard deviation has been shown to be a biased estimator of the population value, and consequently, there is bad news and good news. The bad news is that there are two different formulas, one for the sample standard deviation and one for the population standard deviation. The good news is that statisticians do not often work with a population of numbers. They typically only work with samples and make inferences about the populations from which they were drawn. Therefore, we will only use the sample formula. The two formulas are presented as follows:
Sample Standard Deviation
S= )2 (xi x N1 =

Population
(xi x )2 N

where S (capital English letter S) stands for the sample standard deviation.

Correcting for Bias in the ___________________________ Sample Standard Deviation

Notice that the two formulas only differ in their denominators. The sample formula has N 1 in the denominator, and the population has only

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 78

78

STATISTICS: A GENTLE INTRODUCTION

N. When it was determined that the original formula (containing only N in the denominator) consistently underestimated the population value when applied to samples, the correction 1 was added to correct for the bias. Note that the correction makes the numerator larger, and that makes the value of the sample standard deviation larger (if we divide the numerator by a large number, then it makes the numerator smaller; if we divide the numerator by a smaller number, then that makes the numerator larger). The correction for bias has its greatest effect in smaller samples, for example, dividing by 9 instead of 10. In larger samples, the power of the correction is diminished, yet statisticians still leave the correction in the formula, even in the largest samples.

How the Square Root of x2 Is Almost Equivalent to Taking the Absolute Value of x ____________
As mentioned earlier, statisticians first used the absolute value method of obtaining the average deviation as a measure of variation. However, squaring a number and then taking the square root of that number also removes negative signs while maintaining the value of the distance from the mean for that number. For example, if we have a set of numbers with a mean of 4 and our lowest number in the set is 2, then 2 4 = 2. If we square 2, we get 4, and the square root of 4 = 2. Therefore, we have removed the negative sign and preserved the original value of the distance from the mean. Thus, when we observe the standard deviation formula, we see that the numerator is squared, and we take the square root of the final value. However, if we take a set of numbers, the absolute value method for obtaining the standard deviation and the square root of the squares method will yield similar but not identical results. The square root of the square method has the mathematical property of weighting numbers that are farther from the mean more heavily. Thus, given the previous subset of numbers 28, 42, 48, 59, 63 the absolute value method for standard deviation (without the correction for bias) yielded a value of 10.4, while the value of the square root of the squares method is 12.5.

The Computational Formula for Standard Deviation ______________________________

One other refinement of the standard deviation formula has also been made, and this change makes the standard deviation easier to compute. The following is the computational formula for the sample standard deviation:

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 79

Chapter 3: Statistical Parameters

79

S=

x2

N1

x)2 N

Remember that this computational formula is exactly equal to the theoretical formula presented earlier (the proof of their equality is presented in the appendix). The computational formula is simply easier to compute. The theoretical formula requires going through the entire data three times: once to obtain the mean, once again to subtract the mean from each number in the set, and a third time to square and add the numbers together. Note that on most calculators, x and x2 can be performed at the same time; thus, the set of numbers will only have to be entered in once. Of course, many calculators can obtain the sample standard deviation or the population value with just a single button (after entering all of the data). You may wish to practice your algebra, nonetheless, with the computational formula and check your final answer with the automatic buttons on your calculator afterwards. Later in the course, you will be required to pool standard deviations, and the automatic standard deviation buttons of your calculator will not be of use. Your algebraic skills will be required, so it would be good to practice them now.

________________________________________ The Variance

The variance is a third measure of variation. It has an intimate mathematical relationship with standard deviation. Variance is defined as the average of the square of the deviations of a set of scores from their mean. In other words, we use the same formula as we did for the standard deviation, except that we do not take the square root of the final value. The formulas are presented as follows:

Sample Variance
S2 = )2 (xi x N1

Population
2 = )2 (xi x N

Statisticians frequently talk about the variance of a set of data, and it is an often-used parameter in inferential statistics. However, it has some conceptual drawbacks. One of them is that the formula for variance leaves the units of measurement squared. For example, if we said that the standard deviation for shoe sizes is 2 inches, it would have a clear conceptual meaning. However, imagine if we said the variance for shoe sizes is 4 inches squared. What in the world does inches squared mean for a shoe size?

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 80

80

STATISTICS: A GENTLE INTRODUCTION

This conceptual drawback is one of the reasons that the concept of standard deviation is more popular in descriptive and inferential statistics.

The Sampling Distribution of Means, the Central Limit Theorem, and the Standard Error of the Mean ___________________
A sampling distribution is a theoretical frequency distribution that is based on repeated (a large number of times) sampling of n-sized samples from a given population. If the means for each of these many samples are formed into a frequency distribution, the result is a sampling distribution. In 1810, French mathematician Pierre LaPlace (17491827) formulated the central limit theorem, which postulated that if a population is normally distributed, the sampling distribution of the means will also be normally distributed. However, more important, even if the scores in the population are not normally distributed, the sampling distribution of means will still be normally distributed. The standard deviation of a sampling distribution of means is called the standard error of the mean. As the sample size of the repeated samples becomes larger, the standard error of the mean becomes smaller. Also, as the sample size increases, the sampling distribution approaches a normal distribution. At what size a sample does the sampling distribution approach a normal distribution? Most statisticians have decided upon n = 30. It is important to note, however, if the population from which the sample is drawn is normal, then any size sample will lead to a normally distributed sampling distribution. The characteristics of the sampling distribution serve as an important foundation for hypothesis testing and the statistical tests presented later in this book. However, at this point in a statistics course, it might be important to remember at least this: In inferential statistics, a sample size becomes arbitrarily large at n 30.

The Use of the Standard Deviation for Prediction _____________________________

Pafrutti Tchebysheff (18211894), a Russian mathematician, developed a theorem, which ultimately led to many practical applications of the standard deviation. Tchebysheffs theorem could be applied to samples or populations, and it stated that specific predictions could be made about how many numbers in a set would fall within a standard deviation or standard deviations from the mean. However, the theorem was found to be conservative, and statisticians developed the notion of the empirical rule. The empirical rule holds only for normal distribution or relatively mound-shaped distributions.

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 81

Chapter 3: Statistical Parameters

81

The empirical rule predicts the following: 1. Approximately 68% of all numbers in a set will fall within 1 standard deviation of the mean. 2. Approximately 95% of all numbers in a set will fall within 2 standard deviations of the mean. 3. Approximately 99% of all numbers in a set will fall within 3 standard deviations of the mean. For example, let us return to the ages of the subjects in the dream study previously mentioned: 28, 34, 40, 40, 42, 43, 45, 48, 59, 59, 63, 63, 81, 88 The mean is 52.4. The sample standard deviation computational formula is as follows:
( x)2 N = (733)2 14 = 537, 289 14 13

S= S=

x2

42,287 13

42,287

N1

42,287 38,377.7857 = 13

3,909.2143 = 300.7088 = 17.34 13

Thus, S = 17.34. Now, let us see what predictions the empirical rule will make regarding this mean and standard deviation. - + 1 S = 52.4 + 17.3 = 69.7 1. x - 1 S = 52.4 17.3 = 35.1 x Thus, the empirical rule predicts that approximately 68% of all the numbers will fall within this range of 35.1 years old to 69.7 years old. If we examine the data, we find that 10 of the 14 numbers are within that range, and 10/14 is about 70%. We find, therefore, that the empirical rule was relatively accurate but conservative. - + 2 S = 52.4 + 2 (17.3) = 52.4 + 34.6 = 87.0 2. x - 2 S = 52.4 2 (17.3) = 52.4 34.6 = 17.8 x Inspection of the data reveals that 13 of the 14 numbers in the set fall within two standard deviations of the mean, or approximately 93%. The empirical rule predicted about 95%; thus, it was relatively accurate for these data, but this time it was too liberal a prediction.

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 82

82

STATISTICS: A GENTLE INTRODUCTION

- + 3 S = 52.4 + 3 (17.3) = 52.4 + 51.9 = 104.3 3. x - 3 S = 52.4 3 (17.3) = 52.4 51.9 = 0.5 x All 14 of the 14 total numbers fall within three standard deviations of the mean. The empirical rule predicted 99%, and again we see the rule was relatively accurate and conservative.

Practical Uses of the Empirical Rule: As a Definition of an Outlier _________________________

Some statisticians use the empirical rule to define outliers. For example, a few statisticians define an outlier as a score in the set that falls outside of 3 standard deviations of the mean. However, caution is still urged before deciding to eliminate even these scores from an analysis because although they may be improbable, they still occur.

Practical Uses of the Empirical Rule: Prediction and IQ Tests ______________________________

The empirical rule has great practical significance in the social sciences and other areas. For example, IQ scores (on Wechslers IQ tests) have a theoretical mean of 100 and a standard deviation of 15. Therefore, we can predict with a reasonable degree of accuracy that 68% of a random sample of normal people taking the test should have IQs between 85 and 115. Furthermore, only 5% of this sample should have an IQ below 70 or above 130 because the empirical rule predicted that 95% would fall within two standard deviations of the mean. Because IQ scores are assumed to be normally distributed and both tails of the distribution are symmetrical, we can predict that 2.5% of people will have an IQ less than 70 and 2.5% will have IQs greater than 130. What percentage of people will have IQs greater than 145? An IQ of 145 is exactly three standard deviations above the mean. The empirical rule predicts that 99% should fall within 3 standard deviations of the mean. Therefore, of the 1.0% who fall above 145 or below 55, 0.5% will have IQs above 145.

Some Further Comments _____________________________

The two categories of parameters, measures of central tendency and measures of variation, are important in both simple descriptive statistics and in inferential statistics. As presented, you have seen how parameters from both categories are necessary to describe data. Remember that the purpose of statistics is to summarize numbers clearly and concisely. The parameters, mean and standard deviation, frequently accomplish these two goals, and a parameter from each

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 83

Chapter 3: Statistical Parameters

83

category is necessary to describe data. However, being able to understand the data clearly is the most important goal of statistics. Thus, not always will the mean and standard deviation be the appropriate parameters to describe data. Sometimes, the median will make better sense of the data, and most measures of variability do not make sense for nominal or categorical data.

History Trivia
Fisher to Eels
Ronald A. Fisher (18901962) received an undergraduate degree in astronomy in England. After graduation, he worked as a statistician and taught mathematics. At the age of 29, he was hired at an agricultural experimental station. Part of the lure of the position was that they had gathered approximately 70 years of data on wheat crop yields and weather conditions. The director of the station wanted to see if Fisher could statistically analyze the data and make some conclusions. Fisher kept the position for 14 years. Consequently, modern statistics came to develop some strong theoretical roots in the science of agriculture. Fisher wrote two classic books on statistics, published in 1925 and 1935. He also gave modern statistics two of its three most frequently used statistical tests, t tests and analysis of variance. Later in his career, in 1954, he published an interesting story of a scientific discovery about eels and the standard deviation. The story is as follows: Johannes Schmidt was an ichthyologist (one who studies fish) and biometrician (one who applies mathematical and statistical theory to biology). One of his topics of interest was the number of vertebrae in various species of fish. By establishing means and standard deviations for the number of vertebrae, he was able to differentiate between samples of the same species depending on where they were spawned. In some cases, he could even differentiate between two samples from different parts of a fjord or bay. However, with eels, he found approximately the same mean and same large standard deviation from samples from all over Europe, Iceland, and Egypt. Therefore, he inferred that eels from all these different places had the same breeding ground in the ocean. A research expedition in the Western Atlantic Ocean subsequently confirmed his speculation. In fact, Fisher notes, the expedition found a different species of eel larvae for eels of the Eastern rivers of North America and the Gulf of Mexico.

Key Terms, Symbols, and Definitions

Central limit theoremA mathematical proposition that states that if a populations scores are normally distributed, the sampling distribution of means will also be normally distributed, and if the scores in the population are not normally distributed, the sampling distribution of means will still be normally distributed.

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 84

84

STATISTICS: A GENTLE INTRODUCTION

Empirical ruleA rule of thumb that dictates that approximately 68% of all scores will fall within plus or minus one standard deviation of the mean in a normal distribution, 95% will fall within plus or minus two standard deviations, and 99% will fall within plus or minus three standard deviations of the mean. Equivocal resultsTest results that are unclear, uncertain, or unreadable. Indeterminate resultsA condition when a tests outcome is neither positive nor negative and does not fall between these two conditions. Intermediate resultsA condition when a tests outcome is neither positive nor negative but falls between these two conditions. KlinkersA term used for a datum that is clearly wrong or inappropriate (e.g., a number not on the measuring scale).

-)The arithmetical average of a group of scores. Mean ( x

Measures of central tendencyParameters that measure the center of a frequency distribution. Measures of variationParameters that measure the tendency of scores to be close or far away from the center point. MedianThe center of a distribution of scores, such that half of the scores are above that number, and half of the scores in the distribution are below that number. ModeThe most frequently occurring score. OutliersThese are deviant scores that have been legitimately gathered and are not due to equipment failures, yet they are uncommon scores or they range far from the central tendency of the data set. RangeThe lowest score and the highest score in a distribution. Sampling distributionA theoretical frequency distribution based on repeated sampling of n-sized samples from a given population. If the means of these samples are formed into a frequency distribution, the result is a sampling distribution. Standard deviation ( or SD)A parameter of variability of data about the mean score. Standard error of the meanThe standard deviation of a sampling distribution of means. Unbiased estimatorA sample parameter that neither overestimates nor underestimates the population value. Uninterpretable resultsA condition when a tests outcome is invalid because it has not been given according to its correct instructions or directions. Variance (2)A parameter of variability of data about the mean score, which is the square of the standard deviation.

Chapter 3 Practice Problems

1. Find the mean, median, and mode for each of the following sets of scores: a. b. c. d. 35, 55, 80, 72, 55, 66, 74 110, 115, 102, 102, 107, 102, 108, 110 21, 19, 18, 30, 16, 30 24, 26, 27, 22, 23, 22, 26

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 85

Chapter 3: Statistical Parameters

85

2. To familiarize yourself with the measures of variation, compute the range, standard deviation, and variance for each set of scores. a. 6, .5, .8, .6, .2, .3, .9, .9, .7 b. 9.62, 9.31, 9.15, 10.11, 9.84, 10.78, 9.08, 10.17, 11.23, 12.45 c. 1001, 1253, 1234, 1171, 1125, 1099, 1040, 999, 1090, 1066, 1201, 1356 3. State the differences between klinkers and outliers. How do statisticians deal with each? 4. State the three predictions made by the empirical rule.

Chapter 3 Test Questions

1. Which of the following is NOT true of the mean? a. b. c. d. it is appropriate for normal and skewed distributions it is the same as the arithmetic average it belongs in the category of parameters of central tendency it is considered an unbiased estimator

2. The unbiased aspect of an unbiased estimator indicates that a. it holds for all ethnic groups b. it underestimates the population value with the same tendency as it overestimates the population value c. it works in skewed as well as nonskewed distributions d. all of the above 3. The most important reason that the median is preferred to the mean is a. b. c. d. it is more useful than the mean in inferential statistics it is an unbiased estimator of the population median value it is a more accurate measure of central tendency in highly skewed distributions it is a better measure of variation

4. What is the mean (rounded to two decimal places) of the following set of first-time statistics students ages: 18, 18, 18, 19, 19, 19, 19, 20, 22, 22, 23, 23, 23, 24, 25, 26, 26, 29, 33, 64? a. 24.5 b. 25 c. 24.50 5. What is the median for the previous set of scores? a. b. c. d. 22.5 22 23 23.5

6. What is the mode score, and what is its frequency? a. b. c. d. 18 and 3 23 and 3 19 and 4 4 and 19

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 86

86

STATISTICS: A GENTLE INTRODUCTION

7. What is the standard deviation (rounded to one decimal place) for the previous set of scores? a. b. c. d. 10.12 10.11 10.1 10.2

8. What is the variance (rounded to one decimal place) of the previous set of scores? a. b. c. d. 102.4 102.3 102.368 102

9. What is the range of the previous set of scores? a. b. c. d. 18 64 46 490

10. With respect to the mean and the median values of the previous set of scores, which of the following is true? a. b. c. d. the median is lower than the mean, suggesting a negative skew the median is lower than the mean, suggesting a positive skew the standard deviation is lower than the variance, suggesting a positive skew both b and c are correct

Problems 1115. The following is a list of annual salaries (in dollars) for a group of mid-level managers in a large company.

98,000 65,000 88,000 97,000 93,000 87,000 81,000

115,000 107,000 57,000 85,000 88,000 83,000 76,000

75,000 72,000 88,000 87,000 71,000 75,000 81,000

88,000 71,000 73,000 73,000 65,000 44,000 89,000

11. What is the mean of the salaries (rounded to a whole number)? a. b. c. d. 81,000 81,143 88,000 14,864

12. What is the median of the salaries (rounded to a whole number)? a. b. c. d. 81,000 81,111 88,000 14,864

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 87

Chapter 3: Statistical Parameters 13. What is the mode salary (rounded to a whole number)? a. b. c. d. 81,000 81,111 88,000 14,864

87

14. What is the standard deviation of the salaries (rounded to a whole number)? a. b. c. d. 81,000 81,111 88,000 14,701

15. The top salary in the previous set was $115,000. Would you consider that salary a. a klinker b. an outlier c. neither because it falls well within three standard deviations of the mean Problems 1621. The following is a list of annual advertising expenses (in dollars) for a group of states mental health centers. 1500 600 500 1750 9000 11,000 2350 6000 1250 890 7500 3000 8000 50 1600 800 3800 500 3400 2200

16. What is the mean of the advertising expenses (rounded to a whole number)? a. b. c. d. $3284 $3285 $1975 $3241

17. What is the median of the advertising expenses (rounded to a whole number)? a. b. c. d. $3284 $3285 $1975 $3241

18. What is the mode advertising expense (rounded to a whole number)? a. b. c. d. $500 $3285 $1975 $3241

19. What is the standard deviation of the advertising expenses (rounded to a whole number)? a. b. c. d. $3284 $3285 $1975 $3241

03-Coolidge-4857.qxd

1/2/2006

5:43 PM

Page 88

88

STATISTICS: A GENTLE INTRODUCTION

20. The lowest advertising expense in the previous set was $50. Would you consider that expense a. a klinker b. an outlier c. neither because it falls well within three standard deviations of the mean 21. What is the range of the previous advertising expenses? a. b. c. d. $11,050 $10,050 $10,950 none of the above

22. The most common IQ tests have a standardized mean of 100 and a standard deviation of 15. Based on this information, use the empirical rule to answer the following question: If approximately 348 prospective college students are tested for their IQ, approximately how many will fall within one standard deviation of the mean? a. b. c. d. 345 338 313 237

23. Approximately how many students will have at least an IQ of 100 but not greater than 115? a. b. c. d. 68% 169 174 118

24. Approximately how many students will have IQs greater than 100? a. b. c. d. 118 174 169 237

25. Approximately how many students should have IQs greater than 130? a. b. c. d. 9 18 27 36